Ensemble Data Assimilation of Water Quality Variables

Transcription

1 CAHMDA-DAFOH Joint Seminar, September 6-13, 2014, Austin, TX Ensemble Data Assimilation of Water Quality Variables Kyunghyun Kim*, Lan Joo Park, Minji Park, Changmin Shin, and Joong-Hyuk Min National Institute of Environmental Research, Incheon, Republic of Korea * Corresponding author: matthias@gmail.com 1

2 Algal bloom outbreaks Cyanobacteria bloom outbreaks in major rivers and lakes have been important environmental issues in Korea Photo: 2

3 Algal bloom outbreaks Cyanobacteria bloom outbreaks in major rivers and lakes have been important environmental issues in Korea <Before> Ave. water depth = 2.1m <After> Ave. water depth = 8.5m 8.6m 8.9m 10.4m 8.9m 9.3m 7.9m 7.5m 7.7m 7.4m Photo: 3

4 Effort of WQ degradation prevention Flushing-out is one of active measures to respond major cyanobacteria blooms - As an example, abrupt increase of discharge from Choongju dam was made to flush out cyanobacteria bloom in Lake Paldang (summer 2012)

5 Chl-a (mg/m 3 3 ) ) Efforts of WQ degradation prevention Various scenarios were simulated to determine amount and duration of discharge and actual release was made based on one selected scenario 80.0 Chl-a (Base) Rainfall forecast on Aug. 10 Observed rainfall Chl-a (Scenario I) Chl-a (Scenario II) Chl-a (Scenario III) 39 mm 40 mm 50.0 Chl-a (Observed) 124 mm Paldang dam CJ dam release 8/6 8/7 8/8 8/9 8/10 8/11 8/12 8/13 8/14 8/15 8/16 8/17 Date (2012/8/6 to 8/17)

6 Algal bloom forecast Since January 2012, NIER has been producing 7-days algal bloom forecast for the 16 weir locations in the major rivers in Korea - Forecasting variables: water temp. and Chlorophyll-a level - It will be extended to other WQ variables in the future (e.g., TOC & SS) - Forecasting model: a HSPF-EFDC coupled model developed for the four watersheds - Forecasting report: A 7-days WQ forecast - are officially announced on every Monday and Thursday and circulated to water management agencies in the Han River basin via a dedicated website.

7 A summary of the first two years forecast The RMSE of water temperature forecast for each location tends to increase with lead time but not significantly Variation of the mean RMSE for each river with lead time Variation of RMSE for each location with lead time

8 A summary of the first two years forecast The RMSE of chlorophyll-a forecast for each location tends to increase with lead time but not significantly Variation of the mean RMSE for each river with lead time Variation of RMSE for each location with lead time

9 Reducing Forecast Error : Data assimilation Ensemble Kalman Filter for EFDC model Ensemble representation of EFDC model prediction is created by applying perturbation to HSPF Error Models Point sources Meteorological variables HSPF model prediction based on error models, which are built by comparing model prediction and observation EFDC Non-point sources Ensemble represents all error components associated with HSPF model prediction DA points

10 Study site - Han River watershed

11 Results of Chl-a ensemble simulation with DA

12 Results of Chl-a ensemble simulation with DA

13 Results of Chl-a ensemble simulation with DA

14 How fast does the effect of DA disappear?

15 Correlation among WQ variables Model equations Chl-a PO4 15

16 Results of phosphate ensemble simulation with DA

17 Results of Chl-a ensemble simulation with DA

18 Synthetic experiment Ill-posed ensemble in terms of correlation among WQ variables due to ignorance of spatial correlation of the boundary forcing to EFDC (perturbed HSPF outputs) and lack of consideration on correlation between observed WQ variables Synthetic experiment for well-posed ensemble Perturbed water temperature only and left other WQ variables as single time series in the boundary forcing

19 Synthetic experiment Chl-a PO4

20 Synthetic experiment Chl-a PO4

21 Chl-a NH4 NO3

22 So what is a proper approach? Our approach using error models is wrong. It is very difficult to consider spatial and inter variable correlations in this framework Error Models Point sources Meteorological variables HSPF EFDC Non-point sources Ensemble represents all error components associated with HSPF model prediction DA points

23 So what is a proper approach? In stead, ensemble simulation of the entire watershed is necessary Meteorological variables EFDC Point sources HSPF Non-point sources DA points

24 Thank you National Institute of Environmental Research, Incheon, Republic of Korea 24